As more businesses explore the benefits of cloud computing, network managers will have to increasingly meet the challenges of redesigning their networks to compensate for the unique needs of the cloud.
That's the conclusion drawn from The 2011 Cloud Networking Report, a summary white paper from Ashton Metzler and Associates and Interop that details the three primary challenges facing cloud networking: data center LAN architecture, wide area networking, and network management complexity.
The data center LAN offers a particularly interesting challenge, according to the white paper's author, Dr. Jim Metzler. To this point, LANs within data centers have compensated for the complexity of virtualization by adding multiple layers of control to network management schemes. This was done because each virtualized server contained its own virtualized switch (which Metzler refers to as a "vSwitch"). With potentially thousands of vSwitches in place, the additional layers of management--additional tiers--were added to data center networks.
Metzler argues that the opposite approach is needed, because cloud architectures are different from data centers. True cloud systems depend on automation for cloud management, and thus more server-to-server communication is required. With that kind of communication going on, Metzler reasons in the paper, less layers are needed, not more.
"One approach for improving server-to-server communications is to flatten the data center LAN from the current norm that is either a three or four tier design, to a two tier LAN design consisting of access layer and aggregation/core layer switches," Metzler writes.
Of course, while flattening the layers in the LAN design will help, it still doesn't directly solve the problem of what to do with all of those vSwitches floating around. Here, Metzler offers another approach: edge virtual bridging (EVB).
"With EVB, all the traffic from VMs is sent to the network access switch. If the traffic is destined for a VM on the same physical server, the access switch returns the packets to the server over the same port on which it was received; e.g., a 'hair pin turn,'" Metzler explains. EVB is an IEEE standard that uses Virtual Ethernet Port Aggregators (VEPA) to handle this 180-degree redirection of network traffic, which gives network access switches access and control over an virtual machine traffic.
There are other approaches that can be used to manage the problem of vSwitches, of course. Distributed virtual switching, for instance, separates the control and data aspects of a vSwitch, enables third-party control software to manage the data flows of several vSwitches at the same time.
Whichever approach is used, data center LANs are a big area of improvement for cloud systems. High availability requirements will require the implementation of redundant systems, as well as improvements in Ethernet technology itself, like using a common high-speed Ethernet switching fabric to handle both storage and data traffic.
"This unified fabric offers significant cost savings in multiple areas including converged network adapters on servers and reductions in rack space, power and cooling capacity, cabling, and network management overhead," Metzler writes. "Traditional Ethernet, however, only provides a best effort service. In order to emulate the lossless behavior of a Fibre Channel SAN, Ethernet must be enhanced in such as way that it exhibits lossless behavior."